Monotherapy versus combination therapy for multidrug-resistant Gram-negative infections: Systematic Review and Meta-Analysis

Infections caused by carbapenemase-producing, multidrug-resistant (MDR), or extensively drug-resistant (XDR) Gram-negative bacteria constitute a major therapeutic challenge. Whether combination antibiotic therapy is superior to monotherapy remains unknown. In this systematic review and meta-analysis OVID MEDLINE, EMBASE, PubMed, The Cochrane Library, and Scopus databases were searched for randomized controlled trials (RCTs) and observational studies published by December 2016 comparing mono- with combination antibiotic therapy for infections with carbapenemase-producing, MDR, or XDR Gram-negative bacteria. Mortality and clinical cure rates served as primary and secondary outcome measures, respectively. Of 8847 initially identified studies, 53 studies – covering pneumonia (n = 10 studies), blood stream (n = 15), osteoarticular (n = 1), and mixed infections (n = 27) - were included. 41% (n = 1848) of patients underwent monotherapy, and 59% (n = 2666) combination therapy. In case series/cohort studies (n = 45) mortality was lower with combination- vs. monotherapy (RR 0.83, CI 0.73–0.93, p = 0.002, I2 = 24%). Subgroup analysis revealed lower mortality with combination therapy with at least two in-vitro active antibiotics, in blood stream infections, and carbapenemase-producing Enterobacteriaceae. No mortality difference was seen in case-control studies (n = 6) and RCTs (n = 2). Cure rates did not differ regardless of study type. The two included RCTs had a high and unknown risk of bias, respectively. 16.7% (1/6) of case-control studies and 37.8% (17/45) of cases series/cohort studies were of good quality, whereas quality was poor in the remaining studies. In conclusion, combination antimicrobial therapy of multidrug-resistant Gram-negative bacteria appears to be superior to monotherapy with regard to mortality.

Data extraction and quality assessment. Data extraction from included trials was performed by two of three authors (AS, AW, SK) using a standardized data collection form. In case of missing data, no attemps were made to contact study authors. Following data were extracted from studies: study design, microorganism and type of resistance, disease, antibiotics used in mono-and combination therapy and outcome measures. All-cause mortality and clinical cure rates, as defined in the individual studies, served as primary and secondary outcome measures, respectively. If the endpoints mortality or cure was only reported as probability, data were extracted as proportions.
In the monotherapy group, only patients treated with a single antibiotic drug were included, whereas in the combination therapy group patients treated with two or more antibiotics given simultaneously were includedregardless of in vitro susceptibility. Beta-lactamase inhibitors (i.e. sulbactam, clavulanic acid) were not considered independent antibiotic substances.
Quality assessment of RCTs with regard to method and risk of bias included: sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting, and other sources of bias. Definitions/criteria were derived from the Cochrane Handbook for Systematic Reviews of Interventions 4 . Quality assessment and risk of bias of case-control studies comprised following criteria: selection, comparability, and exposure, according to the Newcastle-Ottawa quality assessment scale for case-control studies 5 . To assess quality and risk of bias of cohort studies, selection, comparability, and outcome was accounted for, using the Newcastle-Ottawa quality assessment scale for cohort studies 5 . Thresholds for converting the Newcastle-Ottawa scales to Agency for Healthcare Research and Quality (AHRQ) standards (good, fair and poor) were as follows: good quality if three or four stars in selection domain and one or two stars in comparability domain and two or three stars in outcome/exposure domain. Fair quality if two stars in selection domain and one or two stars in comparability domain and two or three stars in outcome/exposure domain. Poor quality if zero or one star in selection domain or zero stars in comparability domain or zero or one star in outcome/exposure domain 6 .

Data synthesis and analysis. For Data synthesis and analysis, Review Manager (Version 5.3, The Cochrane
Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) was used. Due to anticipated heterogeneity between studies, random effects models were used for all analyses to obtain a summary estimate (risk ratio (RR)) of the average effect with its 95% confidence interval (CI). P-values ≤ 0.05 were considered statistically significant. Subgroup analyses for various resistance mechanisms, disease entities, antibiotics and quality of studies were performed. Statistical heterogeneity was initially inspected graphically using forest plots. The degree of heterogeneity was quantified by using the I 2 statistic with a threshold of I 2 > 60% for high heterogeneity. Publication bias was investigated using a funnel plot in which the standard error of the effect estimate of each study was plotted against the estimate.

Results
Study selection. Figure 1 shows the study selection process according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses for Protocols (PRISMA). The initial search yielded 8847 references. After screening titles and abstracts and exclusion of duplicates, 182 articles were selected for full-text screening, after which 53 studies were retained for the analysis. Included studies are depicted in Supplementary Table 2. Addressed infections ranged from pneumonia (n = 10 studies) 7-16 , blood stream infections (n = 15 studies) 17-31 , osteoarticular infections (n = 1 study) 32 , and studies comprising mixed forms of infections (n = 27 studies) 33 case series and cohort studies. Meta-analysis of all case series and cohort studies included 1813 patients (56.7%) with combination therapy and 1383 patients (43.3%) with monotherapy and demonstrated a lower mortality risk in the combination therapy group (RR: 0.83, 95% CI 0.73-0.93, p = 0.002). Heterogeneity was low (I 2 = 24%, p = 0.1).
Subgroup analysis of various diseases revealed an advantage of combination therapy when applied in bloodstream infections (RR 0.78, 95% CI 0.63-0.96; p = 0.02; I 2 = 27%, p = 0.17), whereas no difference could be seen in pneumonia or studies including mixed forms of infections. In studies that only included bloodstream infections (n = 14), stratified according to bacteria and resistance mechanism, combination therapy was superior in the group of carbapenemase-producing Enterobacteriaceae (RR: 0.61, CI 95% 0.45-0.85, p = 0.003; I 2 = 21%, p = 0.26), with no difference in MDR/XDR Acinetobacter baumanii and mixed MDR/XDR Gram-negative bacteria ( Supplementary Fig. 1). Subgroup analysis including only infections due to carbapenemase-producing Enterobacteriaceae, stratified by different disease types, again showed lower mortality rates in bloodstream infections (RR 0.61, CI 0.45-0.85, p = 0.003; I 2 = 21%, p = 0.26) with no difference in the non-bloodstream infection group (comprising studies with mixed forms of infections) ( Supplementary Fig. 2).
Subgroup analysis of different treatment strategies did not reveal a distinct combination therapy with a superior outcome compared to monotherapy. Analysis of the 12 studies with combination therapies that contained only one in-vitro active substance did not result in better outcomes against monotherapy, nor did subgroup analyses stratified according to bacteria and diseases. Polymyxin combinations performed better than polymyxin monotherapy (n = 1; RR 0.63, CI 95% 0.41-0.96, p = 0.03). Colistin combinations performed better than monotherapies with various antibiotics (n = 1; RR 0.38, CI 0.22-0.65, p = 0.0005).
Subgroup analysis of good and poor quality studies revealed a significant lower mortality risk with combination antibiotic therapy versus monotherapy in good quality studies (RR: 0.78, 95% CI 0.69-0.88, p = 0.0001, I 2 = 4%, p = 0.4). Among studies with poor quality, no difference in mortality in the combination therapy vs. the monotherapy group was detected. With regard to the different diseases, a significant difference of combination antibiotic therapy was seen in good quality studies of blood stream infections (RR: 0.75, 95% CI 0.61-0.93, p = 0.009, I 2 = 17%, p = 0.3) and good quality studies of mixed infections (RR: 0.77, 95% CI 0.64-0.92, p = 0.004, I 2 = 5%, p = 0.38), whereas no advantage of combination therapy could be shown in poor quality studies. Subgroup analysis of different bacteria and quality of studies revealed a significant lower mortality with case-control studies. Meta-analysis of case-control studies comprising 142 patients (50.4%) with combination therapy and 140 patients (49.6%) with monotherapy did not show a significant difference in mortality between combination therapy and monotherapy. Neither did subgroup analyses of case-control studies comparing different bacteria/resistance mechanisms, diseases and treatments show a difference in mortality between the two groups. In all included case-control studies, either only one in vitro active substance was used in the combination group or it was not clearly stated whether there were several in-vitro active substances in the combination group. case series and cohort studies. Meta-analysis of case series and cohort studies with a total of 958 patients (67.0%) with combination therapy and 472 patients (33.0%) with monotherapy, did not show a difference in clinical cure rates between mono-and combination therapy (Fig. 4).
Subgroup analysis of different bacteria, resistance mechanisms and quality of studies did not reveal a difference between mono-and combination therapy. A trend to higher cure rates in combination therapy was seen in MDR Pseudomonas aeruginosa, MDR Acinetobacter baumanii and carbapenemase-producing Enterobacteriaceae.
Subgroup analysis of different diseases revealed a difference between mono-and combination therapy in osteoarticular infection (n = 1; RR 0.43, CI 0.21-0.89, p = 0.02) but no difference in other types of infections (pneumonia, bloodstream infections, mixed forms of infection).
No difference in cure rates was seen with different antibiotic therapies, regardless of the number of substances with in-vitro activity in combination treatment groups (Fig. 5).
case-control studies. Meta-analysis of case-control studies, including 125 patients (53.0%) with mono-and 111 patients (47.0%) with combination therapy, did not show any difference of monotherapy vs. combination therapy with regard to clinical cure rates. Subgroup analyses looking at different bacteria/resistance mechanisms, diseases (pneumonia, mixed) and treatments did not show any difference in clinical cure rates. www.nature.com/scientificreports www.nature.com/scientificreports/ Study quality. The quality of included studies was variable. Two of the included studies were RCTs, one of which was un-blinded and the other one did not report on blinding leading to potential performance bias. One RCT showed an overall high risk of bias and the second one an overall unknown risk of bias (Supplementary Table 3). Only one out of six included case-control studies (16.7%) and 17 out of 45 (37.8%) included cases series and cohort studies showed an overall good quality with good comparability of cases and controls. The remaining studies were prone to confounding factors resulting in poor quality (Supplementary Table 4 and Supplementary  Table 5). Funnel plots did not suggest publication bias (data not shown).

Discussion
In this large meta-analysis including 53 studies with a total of 4514 patients, overall mortality in case series and cohort studies was lower in patients treated with antibiotic combination therapy. Specifically, lower mortality rates were detected in patients with bloodstream infections and in infections caused by carbapenemase-producing Enterobacteriacae if combination therapy entailed at least two substances with in-vitro activity. There was no difference in mortality rates if combination therapy comprised only one in-vitro active substance, in RCTs (2 studies; several in-vitro active substances in combination therapy) and case-control studies (5 studies; only one in-vitro active substance in combination therapy). These effects were confirmed when restricting our analyses to studies of good quality.
Eight 19,20,24,25,27,29,31 out of fourteen case series and cohort studies (57%) that included patients with bloodstream infections were due to carbapenemase-producing Enterobacteriaceae. These studies also accounted for 53% (8/15) of all included case series and cohort studies due to carbapenemase-producing Enterobacteriaceae. The mortality rate of patients with bloodstream infections with bacteria other than carbapenemase-producing Enterobacteriaceae did not differ between combination and monotherapy. Likewise, non-bloodstream infections by carbapenemase-producing Enterobacteriaceae did not show a superior effect of combination therapy. Thus, it is likely that the superior effect of combination therapy with infections caused by carbapenemase-producing Enterobacteriaceae and with bloodstream infections is actually one and the same.
Our results are in line with published results by Zusman et al. 60 , who showed lower mortality with polymyxin/ carbapenem or polymyxin/tigecycline, aminoglycosides or fosfomycin combination therapy compared to polymyxin monotherapy in carbapenem-resistant bacteria and could also show a benefit of combined treatment in bloodstream infections. In contrast to our meta-analysis, Zusman et al. included carbapenem-resistant bacteria in general (i.e. also including carbapenem resistance due to porin loss), and regarded betalactamase inhibitors (i.e. sulbactam) as separate antibiotic substances. Furthermore, Zusman et al. only included carbapenem-resistant bacteria and only polymyxin mono-and combination therapy, whereas our analysis comprised different multiresistant bacteria and antimicrobials.
In theory, several facts argue against the application of combination therapy: the possible rise in resistance rates due to an overall increase in selection pressure as a result of a greater release of antibiotics into the environment 61 , higher rates of adverse effects (such as nephro-and ototoxicity due to colistin or aminoglycosides) 61 , an increase in Clostridium difficile-associated infections, fungal infections 3 , higher costs 61   www.nature.com/scientificreports www.nature.com/scientificreports/ monotherapy 62 , lower risk of resistance development (e.g. against colistin) 1 , lower doses and shorter treatment duration 61 .
Observational study designs are prone to several limitations, mainly due to unmeasured confounding factors and other risks of bias. The likelihood of inappropriate empirical antibiotic therapy is higher in the monotherapy group and the time to start of appropriate monotherapy after results from susceptibilty testing are known may be longer compared to the combination therapy group, which may result in a bias in favour of combination therapy 3 . Information about the dosing of colistin was only reported in a minority of the included studies 12,27,29,34 . As guidance has changed in recent years and a loading dose is now generally recommended, colistin could have been underdosed in a significant proportion of studies, which may constitute a bias against colistin monotherapy 3 . Furthermore, there is a chance of selection bias in favour of combination therapy, as patients are more likely to receive colistin or tigecyclin monotherapy if bacteria are resistant to all other antibiotics as compared to situations where bacteria are still susceptible to carbapenems, for example, or in patients who suffer from polymicrobial infections with resistant as well as susceptible strains 3 . Such an effect could result in a perceived lower overall mortality in the combination treatment group at least partly due to a subgroup of carbapenem susceptible Gram-negative bacteria 20,29,56 . Moreover, combination therapy is more likely prescribed to patients with polymicrobial infections in which one does not know for sure whether the resistant bacterium is the main causative agent 3 . Against the latter argues the fact that an advantage of combination therapy could be shown in bloodstream infections, where one can almost be certain of the main causative agent, in analogy to Zusman and colleagues 60 . Finally, one could argue that combination therapy is more likely to be prescribed in more serious infections, which would bias the results in favour of monotherapy.
Due to numerous limitations and biases, which is reflected in the quality assessment of included studies, the results of this metaanalysis have to be interpreted with caution. However, by inclusion of a wide variety of cohort studies and case series, including a heterogeneous group of infectious diseases, bacteria, and treatments, we tried to minimize existing biases and limitations and allow for a valid comparison of combination treatment with monotherapy. Moreover, the favourable effect of combination therapy in bloodstream infections due to carbapenmase-producing Enterobacteriaceae has recently been confirmed in a large retrospective cohort for the subset of patients with severe infections 63 . With regard to RCTs and case-control studies, the validity of the results of our meta-analysis of RCTs and case-control studies is very limited due to the low number and limited power of the included studies.
To reach a final conclusion as to the possible advantage of combination therapy versus monotherapy, well designed and sufficiently powered RCTs are needed. Two RCT's currently being undertaken could bring some more clarity in this contentious issue 64,65 . Such trials, however, will always have the problem to be prone to the conclusion that their findings can only be applied to the population under study. Only pragmatic trial designs will help resolve this issue. For the time being and taking into account limiting factors of RCTs, such as problems in reaching sufficient power in the setting of uncommon diseases as well as costs, other forms of study (cohort studies, case series, case-control studies) provide valid and valuable data on the best evidence-based treatment 66 . Individual patient-data meta-analyses may add further insights when it proves to be difficult to obtain large enough sample sizes for certain study questions.
Based on our results, combination therapy with two antimicrobials that demonstrate in vitro activity should be administered in patients with bloodstream infections due to carbapenemase-producing Enterobacteriaceae.